# Benchmark capture standards This is the media contract for Coaster Maps benchmark evidence. A capture is eligible for blind voting only when the subject is both identifiable and shown at a useful presentation distance. A technically valid WebM or PNG is not enough if the camera points at the wrong part of the experience. ## Infinite Coaster scope lock The only Infinite Coaster comparison published on CoasterMaps.com is `infinite-coaster-v5`. Only these four registered runs are eligible: - `infinite-coaster-v5-sol-ultra-reference` - `infinite-coaster-v5-grok-4-5-high` - `infinite-coaster-v5-grok-4-5-medium` - `infinite-coaster-v5-grok-4-5-low` Historical Infinite Coaster v1-v4 candidate directories may remain in the analysis workspace for provenance, but they are out of scope. Do not capture, stage, publish, or compare their media. Every v5 still, clip, manifest item, and website submission must retain the `infinite-coaster-v5` benchmark ID and one of the four run IDs above. ## Required presentation - Capture at the registered desktop viewport and keep the complete subject in frame. Use a clean canvas: no HUD, browser chrome, debug text, pointer, or capture controls. - Use a stable camera with a deliberate composition. The subject should occupy enough of the frame to be judged without guessing what the viewer is meant to inspect. - Ordinary motion clips run for 12 seconds. Moving subjects that do not have a trustworthy cycle boundary run for 20 seconds. A ride-cycle duration is used only when the instrumentation proves a complete cycle. If a fast ride would return to its station inside the twenty-second window, register an explicit action-window override of at least four seconds and end while the train is still in the primary element; never submit the station/brake tail. - Use WebM video at 30 fps with a readable first frame and continuous motion. Reject tiny, empty, frozen, corrupted, or unexpectedly short files. - Still images and video clips for the same item must use the same subject, name, category, and run identity. ## Subject framing contract ### Entrance gate Show the actual park entrance from the public approach: gate structure, banner/logo, pylons or ticketing context, and the opening should be legible. Use a front or three-quarter view that makes the entrance unmistakable. Avoid a distant aerial where the gate is a few pixels wide, a view from inside the park looking back, a side/back view, foliage or UI covering the sign, and an empty path with the gate outside the frame. ### Park overview Show the park as a coherent place: central circulation, multiple attractions, and enough terrain or skyline to establish scale. Keep the camera high enough to read the layout but low enough that the attractions remain recognizable. Avoid a title screen, a featureless sky/ground frame, or an overview that is so wide that no attraction can be judged. ### Roller coaster exterior / train Show the train visibly moving through the primary attraction: a drop, launch, loop, corkscrew, airtime hill, banked turn, or other high-interest element. Orbit beside or follow behind the train so the track geometry and motion are both readable. The subject should not begin or spend the clip in the station, on a slow lift hill, or on a brake/queue segment. If the capture system starts from the station, it must wait to begin recording until the train is running outside the station/lift zone. Avoid an empty track, a train hidden behind terrain, a camera so far away that the train disappears, a camera attached to scenery, a view that only shows the lift chain, and a clip whose most important action occurs before the first recorded frame. For a short action window, reject any final frames that have already returned to the station. ### Coaster rider POV Show forward motion from the rider seat through the exciting portion of the course. The track, horizon, and major element should remain readable without violent camera jitter or a HUD obscuring the view. Avoid station boarding, a long lift-only opening, a blank sky/ground view, severe clipping through the train, and excessive look-around that hides the ride. ### Flat rides Show the complete ride silhouette and the motion that identifies it: wheel rotation, carousel rotation, pirate-ship swing, or drop-tower movement. Keep the camera close enough to show the ride body and far enough to avoid clipping. Avoid a static ride between cycles, a view blocked by a roof or tree, and a composition that crops the ride’s defining mechanism. ### Shop / storefront Show the front face of the shop, including its sign, awning/window, and enough of the counter or facade to identify it as a storefront. Use a close, front or slightly three-quarter view; zoom in until the sign and facade are genuinely judgeable. Avoid a back or side wall, a distant orbit, a view aimed at the roof, an empty interior, or people/props hiding the storefront sign. ### Entertainer / staff Show the entertainer or staff figure close-up, unobstructed, and large enough to identify the role and visible prop/action. Leave a little headroom and keep the figure against a readable background. Avoid a crowd blocking the figure, a distant wide shot, severe depth/fog occlusion, a cropped head/prop, or a view where the performer is only a colored dot. ### Game, station, path, guests, and scenery Frame the named subject first, then add enough surrounding context to explain what it is. For a game, show the stall and target area; for a station, show the platform, canopy, sign, and queue; for guests/scenery, show the authored subject clearly rather than using an unrelated wide shot. ## Infinite Coaster v5 category contract The published v5 visual catalog currently contains these 30 blind-voting categories. Keep the category ID, still, clip, and run identity aligned; do not substitute a generic ride view for a named subject. - **Ride presentation:** `coaster-overview`, `coaster-pov`, `track-quality`, `coaster-train`, `environment`, `follow-camera`, `major-drop`, `launch`, `camelback-airtime-hill`, `terrain-following-sequence`, `selected-inversion`. - **Cameras and seats:** `track-pov`, `front-row-rider-pov`, `alternate-seat-pov`. - **Train and operation:** `riders`, `station-dispatch`. - **Desert setting:** `red-rock-formations`, `juniper-tree`, `saguaro-cactus`, `desert-bighorn-sheep`, `waterfall-and-flume`, `terrain-cut`. - **Mining world:** `mining-depot-station`, `mining-settlement`, `stamp-mill`, `timber-trestle`, `mineshaft-experience-a`, `mineshaft-experience-b`. - **Spectacle and identity:** `spectacle-event`, `signature-feature`. - **Ride elements:** `major-drop`, `launch`, `camelback-airtime-hill`, `terrain-following-sequence`, and `selected-inversion`. Use a backed-off side or three-quarter observer view that shows the train moving through the named element, with enough track, supports, terrain, and horizon to judge the transition. In particular, frame the complete inversion and its approach/exit; do not use a distant train-follow shot or a sky-dominated POV. Reject station, lift-only, brake-run, empty track, or late-return footage. - **Cameras and seats:** `track-pov`, `front-row-rider-pov`, `alternate-seat-pov`, and `follow-camera`. Make the camera position obvious: forward track for POVs, the train and terrain for follow camera, and a stable centered view without HUD, severe clipping, or uncontrolled look-around. - **Train and operation:** `riders` and `station-dispatch`. Show visible riders or a readable dispatch sequence with the train, platform, canopy, queue, and motion context. Do not use an empty train, a blocked platform, or a static station hold. - **Desert setting:** `red-rock-formations`, `juniper-tree`, `saguaro-cactus`, `desert-bighorn-sheep`, `waterfall-and-flume`, and `terrain-cut`. Use a subject-centered close orbit or stable three-quarter view for vegetation, wildlife, and water; the coaster does not need to be visible. Fill enough of the frame with the named subject to identify it, while retaining authored terrain and scale cues. If the bighorn is absent or unreadable after the standardized encounter window, preserve the omission and treat it as the lowest outcome against any valid visible encounter; never substitute a generic animal or a scenic shot. Reject tiny distant props, fog-obscured subjects, terrain intersections, and camera views dominated by unrelated ground. - **Mining world:** `mining-depot-station`, `mining-settlement`, `stamp-mill`, `timber-trestle`, `mineshaft-experience-a`, and `mineshaft-experience-b`. Use a close external orbit or three-quarter inspection for structures so the complete facade, machinery, supports, and material detail are readable. Each mine clip must show the portal approach, a practically lit interior, and the outdoor exit in one continuous sequence. Reject empty approaches, back-of-building views, unlit black interiors, and clips where terrain hides the named subject. - **Spectacle and identity:** `spectacle-event` and `signature-feature`. Use a wide but subject-centered observer position or slow orbit that shows the complete active event/feature and its mining-world context. The train is included only when it explains the interaction. Reject a setup frame, inactive state, unrelated scenery, or an effect that obscures the subject. The v5 protocol also requires separate objective continuity and startup evidence; those measurements are not substitutes for the 30 visual voting pairs. The per-category `qualityStandard` in `assets/infinite-coaster-v5-rating-categories.json` is the machine-readable acceptance text used by the catalog validator. ## QA and rejection checklist Before publication, inspect the first frame, a middle frame, and the final frame of every clip and the full-resolution still. Re-capture when any answer is no: 1. Is the named subject visible immediately and still visible at the end? 2. Is the camera aimed at the requested face/element rather than its back, side, station, or lift? 3. Is the subject large enough to judge and free of major occlusion? 4. Does motion begin in the clip’s useful action window rather than after a long setup? 5. Is the image free of HUD/debug/browser artifacts and capture failures? 6. Does the file meet the duration, size, codec, and manifest contract? When a clip fails a content check, changing its label or accepting the file is not a fix. Adjust the camera path or the capture start condition and record a replacement. The capture manifest should preserve the final approved framing parameters so later runs can reproduce the standard.